Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Speaker Deck
PRO
Sign in
Sign up for free
Pythonistaに憧れた分析屋の奮闘記
kanan
June 09, 2017
Programming
0
970
Pythonistaに憧れた分析屋の奮闘記
kanan
June 09, 2017
Tweet
Share
More Decks by kanan
See All by kanan
Python入門_PyLadiesTokyo2021/08/29
kanan
0
150
コトハジメ的Python入門_WiDS広島
kanan
0
60
予測モデルがポンコツになった日_PyLadiesTokyo10May2020-LT
kanan
0
240
Python入門(PyLadies Caravan in Hokkaido)
kanan
3
190
【PyLadies Caravan】大人のためのPython入門 in 愛媛
kanan
0
290
Python入門(PyLadies Caravan in Aichi)
kanan
0
230
PyCon2019体験記-case:KANAN-
kanan
1
560
大人のためのPython入門in京都.pdf
kanan
0
210
Other Decks in Programming
See All in Programming
Modern Android Developer ~ 안내서
pluu
1
660
クックパッドマートの失敗したデータ設計 Before / After 大放出
mokuzon
0
180
Scrum Fest Osaka 2022/5年で200人になったスタートアップの アジャイル開発の歴史とリアル
atamaplus
1
970
ES2022の新機能
smt7174
0
270
即、New Relic / New Relic NOW!
uzulla
0
340
Meet Swift Regex
usamik26
0
370
What's new in Android development tools まとめ
mkeeda
0
410
VisualProgramming_GoogleHome_LINE
nearmugi
1
240
Power Automateドリブンのチームマネジメント
hanaseleb
0
200
git on intellij
hiroto_kitamura
0
170
I/O Extended 2022 in Android ~ Whats new in Android development tools
pluu
0
570
Let's keep Commodore 64 alive for the next 40 years
mehowte
1
110
Featured
See All Featured
The MySQL Ecosystem @ GitHub 2015
samlambert
238
11k
Building an army of robots
kneath
299
40k
Support Driven Design
roundedbygravity
86
8.5k
Designing Dashboards & Data Visualisations in Web Apps
destraynor
224
49k
What the flash - Photography Introduction
edds
62
10k
Six Lessons from altMBA
skipperchong
14
1.4k
Documentation Writing (for coders)
carmenhchung
48
2.6k
A Tale of Four Properties
chriscoyier
149
21k
Making Projects Easy
brettharned
98
4.3k
Visualizing Your Data: Incorporating Mongo into Loggly Infrastructure
mongodb
29
4.3k
Distributed Sagas: A Protocol for Coordinating Microservices
caitiem20
315
19k
Fight the Zombie Pattern Library - RWD Summit 2016
marcelosomers
226
15k
Transcript
PythonistaʹಌΕͨੳͷฃಆه @kanan* 2017.06.07 ΈΜͳͷPythonษڧձ#25
ࣗݾհ ˎKANANʢ͔ͳΜʣ ɹɹcompass-ID @kanan* ɹɹTwitter-ID @Addition_quince ˎ͓ࣄ ɹɹSIerاۀͰσʔλαΠΤϯεؔ࿈ۀΛͬͯ·͢ ɹɹݴޠSAS ˎPyLadiesʹ2016.ळ͔Βͪΐͪ͜ΐ͜ࢀՃ
ˎझຯԅɻ͓ञ͕େ͖ɻ
ࠓͷ͓ͳ͠ ˎϓϩάϥϛϯάʹڵຯ͕ͳ͔ͬͨੳ͕ ˎεΩϧΞοϓͨͯ͘͠PythonʹखΛग़ͦ͏ͱܾҙ͠ ˎ৺ંΕͳ͕Βฃಆͨ͠10ϲ݄ؒͷ͓Ͱ͢ PythonistaʹಌΕͨੳͷฃಆه Pythonista Lv0 → Lv1.5 ※͖ͬͱੌ͍ਓLv
20͘Β͍
Pythonͱͷग़ձ͍ σʔλੳܥɹˠɹPython Α͋͘Δͭ
͜Ε·Ͱͷϓϩάϥϛϯάྺ 2013 2014 2015 2016 2012 2011 େֶ γεςϜΤϯδχΞ σʔλαΠΤϯςΟετ
ˎੲϓϩάϥϛϯάͯͨͭ͠Γͩͬͨ ˎࣄͰϚωδϝϯτ͕ଟ͘ͳΓɺ৮Βͳ͍ۭനͷظؒ ˎSASݴޠͱ͍͍ͳ͕ΒπʔϧͬΆ͍ ϓϩάϥϛϯάɹʹɹࣄ
͜Ε·Ͱͷϓϩάϥϛϯάྺ ͪΌΜͱϓϩάϥϛϯάͬͯͳ͍
PythonσϏϡʔΛܾҙ PyLevel 0 20169݄ ˎҟ༷ͳ΄ͲͷΔؾ ˎΔੳͷࣝ
ͦͷ̍िؒޙɾɾɾ PyLevel 0 20169݄ ͦͷ̎ ˎΠϯετʔϧํ๏ݕࡧ͢Δͱ ɹɹΓํ͕ຯʹҧ͏ ˎHomebrewʁpyenvʁφχιϨ ˎPython3ΛೖΕͨͷʹ ɹɹόʔδϣϯ͕Python2ͬͯͳΔ
ˎ.bash_plofileͳ͍͠ʂ ˎPython͡ΊΔͨΊʹങͬͨ ɹɹMACͷ͍ํ͕Θ͔Βͳ͍ Πϯετʔϧ Ͱ͖ͳ͍ʂ
ܸɹ
ͦΜͳ࣌ɺPyLadiesTokyoͱग़ձ͏
PyLadies TokyoͰSTEP UP PyLadies Tokyo ळ߹॓ 2016 [2016.10.8-10] PyLadies Tokyo
Meet Up #16 [2016.11.27] PyLadies Tokyo Meet Up #17 [2016.12.11] PyLadies Tokyo Meet Up #20 [2016.03.25] *ڥߏங,AnacondaͱJupyterNotebookͷ͍ํΛֶͿ *ڝٕϓϩάϥϛϯάͰPythonͷॻ͖ํΛֶͿ *ϚΠίϯϘʔυ(STM32 Discovery)ʹMicro PythonͰLνΧ *WebεΫϨΠϐϯάΛͬͯΈΔ
ݸਓతʹσʔλੳपลͰษڧ ˎσʔλੳܥ ɹɹɹ1) titanicੜଘऀ༧ଌ ɹɹɹ2) ΞϝϦΧͷՈͷച٫Ձ֨༧ଌ ɹɹɹ3) ίϯϏχͷച্͛༧ଌ ˎͦͷଞ ɹɹɹ1)
खॻ͖ࣈͷೝࣝ ɹɹɹ2) RaspberryPi3ͰPythonͬͯΈΔ 2017.3 ʙ2017.4
PyLevel 1.0 20174݄ ˎΞϧΰϦζϜָ͍͠ ˎԿ͔Ͱ͖Δͱخ͍͠ ˎExcelSASͰ ɹͬͯͨࣄ͕PythonͰ ˎͬͱ৭ʑͬͯΈ͍ͨ Δؾ෮׆ɻͬͱೖऀϨϕϧʹ
νϟϨϯδ ͕ࣗڵຯ͕͋ΔςʔϚͰ Γ͍ͨͱࢥͬͨͷΛΖ͏
ΟεΩʔͰػցֶशʹઓ ʲͳΜͰΟεΩʔʁʳ ɹˎͱΓ͋͑ͣࢲ͕͖ ɹˎϫΠϯΈ͍ͨʹฑ͕Ռͯ͠ͳ͘ଟ͍Θ͚͡Όͳ͍ ɹˎւ֎ͷϏʔϧΈ͍ͨʹຯ߳Γͷಛ͕͖ͬΓͯ͠Δ Ϙτϧͷϥϕϧ͔ΒฑΛผ ը૾ೝࣝ WEBͷใ͔Β֤ฑͷಛΛநग़ ςΩετղੳ ͖ͳฑ͔ΒͨͿΜ͓ޱʹ߹͏ͷΛਪન
Ϩίϝϯυ
Anaconda Continuum Analyticsࣾఏڙͷ σʔλੳͰΑ͘ར༻͞ΕΔϥΠϒϥϦ܊Λ ҰׅΠϯετʔϧͰ͖ͪΌ͏ύοέʔδ ΟεΩʔͰػցֶशʹઓ Jupyter Notebook ϒϥβͰಈ࡞͢Δରܕ࣮ߦڥ ίʔυهड़ͱ࣮ߦɺίϝϯτૠೖ͕Ͱ͖ɺ
݁Ռͷอଘڞ༗ʹศར AnacondaೖΕΔͱσϑΥϧτͰೖͬͯΔ ɹPythonҎ֎ʹR, node.js, RubyෳݴޠʹରԠ ʲ࣮ߦڥʳ
ΟεΩʔͰػցֶशʹઓ Ϙτϧͷϥϕϧ͔ΒฑΛผ ը૾ೝࣝ ̍ʣҰੜݒ໋σʔλΛूΊΔ ̎ʣskimageΛͬͯHOGಛྔΛऔಘ͢Δ ̏ʣֶशσʔλͱςετσʔλʹׂ ̐ʣsklearnΛͬͯCodeBookΛ࡞͠BoVWʹม ̑ʣֶशͱςετ ▪༻ͨ͠ϥΠϒϥϦ ɹɹskimage,
matplotlib, sklearn, glob, os
ΟεΩʔͰػցֶशʹઓ Ϙτϧͷϥϕϧ͔ΒฑΛผ ը૾ೝࣝ ̍ʣҰੜݒ໋σʔλΛूΊΔ 899ຕ ࣗͷ͖ͳ11ฑ͚ͩͰ৺ંΕͯఘΊΔ
ΟεΩʔͰػցֶशʹઓ Ϙτϧͷϥϕϧ͔ΒฑΛผ ը૾ೝࣝ ̎ʣskimageΛͬͯHOGಛྔΛऔಘ͢Δ HOG(Histograms of Oriented Gradients) ɹہॴྖҬ (ηϧ)
ͷًͷޯํΛώετάϥϜԽ 1.ը૾ΛదͳαΠζʹϦαΠζ͠ɺάϨΠεέʔϧͰಡΈࠐΉ 2.֤pixelͷً͔ΒޯڧͱޯํΛٻΊΔ 3.ηϧྖҬ͝ͱʹώετάϥϜΛٻΊΔʢࠓճ8×8ϐΫηϧʣ 4.ϒϩοΫ͝ͱʹਖ਼نԽ͠ɺಛྔΛநग़͢Δʢࠓճ3×3ηϧʣ
ΟεΩʔͰػցֶशʹઓ Ϙτϧͷϥϕϧ͔ΒฑΛผ ը૾ೝࣝ ̎ʣskimageΛͬͯHOGಛྔΛऔಘ͢Δ
ΟεΩʔͰػցֶशʹઓ Ϙτϧͷϥϕϧ͔ΒฑΛผ ը૾ೝࣝ ̐ʣsklearnΛͬͯCodeBookΛ࡞͠BoVWʹม BoVW(Bag-of-Visual-Words)ܗࣜ ɹը૾ͷಛྔΛϕΫτϧԽ͠ώετάϥϜʹͨ͠ͷ ɾɾɾɾ ɾɾɾ ɾɾɾ Visual-word
vectors Codebook (දύλʔϯͷϦετʣ ç ç ç ç
ΟεΩʔͰػցֶशʹઓ Ϙτϧͷϥϕϧ͔ΒฑΛผ ը૾ೝࣝ ̑ʣֶशͱςετ ը૾σʔλͱਖ਼ղͷϥϕϧΛֶशͤ͞Δɹ※SVMʢαϙʔτϕΫλϚγϯʣΛ࠾༻ άϦουαʔνͰϋΠύʔύϥϝʔλνϡʔχϯά
ΟεΩʔͰػցֶशʹઓ Ϙτϧͷϥϕϧ͔ΒฑΛผ ը૾ೝࣝ ਖ਼ ɹ72% ςετ݁Ռ
ΟεΩʔͰػցֶशʹઓ Ϙτϧͷϥϕϧ͔ΒฑΛผ ը૾ೝࣝ ͬͯΈͨ݁Ռ ▪ը૾ͷલॲཧ෦ ɹɹɾࠓճϥϕϧΞοϓͷը૾͔ΓΛ༻ͨ͠ͷͰɺ ɹɹɹҾ͖ͷࣸਅʹରԠͰ͖ͳ͍ʢͬͯΈͨΒյ໓తʣ ɹɹɾഎܠͱ͔ະߟྀͷ·· ɹɹɾOpenCVʁͳʹͦΕ ▪ֶश෦
ɹɹɾಛྔϕʔε(BoVW)ͰͷֶशΛ࠾༻͚ͨ͠Ͳɺ ɹɹɹσΟʔϓϥʔχϯά͏ͱͬͱਫ਼͕͋Δͷ͔ͳ
WEBͷใ͔Β֤ฑͷಛΛநग़ ςΩετղੳ ΟεΩʔͰػցֶशʹઓ ̍ʣWebαΠτ͔Β֤ฑͷઆ໌ίϝϯτΛεΫϨΠϐϯά ̎ʣmecabͰ୯ޠʹׂ͠ɺܗ༰ࢺͱ෭ࢺ͚ͩऔΓग़͢ ̏ʣग़ݱͨ͠୯ޠΛ·ͱΊɺTFIDFܭࢉͰಛతͳޠΛಛఆ ̐ʣ୯ޠΛͬͯࣅ͍ͯΔฑΛΫϥελϦϯά
WEBͷใ͔Β֤ฑͷಛΛநग़ ςΩετղੳ ΟεΩʔͰػցֶशʹઓ ̍ʣWebαΠτ͔Β֤ฑͷઆ໌ίϝϯτΛεΫϨΠϐϯά beautiful soupΛ༻ ɹɹᶃHTMLऔಘ ɹɹᶄ΄͍͠ใͷ෦ͷλάΛݟ͚ͭΔ ɹɹᶅߏղੳͯ͠ಛఆ෦Λநग़ ɹɹᶆσʔλϑϨʔϜʹม͢Δ
WEBͷใ͔Β֤ฑͷಛΛநग़ ςΩετղੳ ΟεΩʔͰػցֶशʹઓ ̍ʣWebαΠτ͔Β֤ฑͷઆ໌ίϝϯτΛεΫϨΠϐϯά beautiful soupΛ༻ ɹɹᶃHTMLऔಘ ɹɹᶄ΄͍͠ใͷ෦ͷλάΛݟ͚ͭΔ ɹɹᶅߏղੳͯ͠ಛఆ෦Λநग़ ɹɹᶆσʔλϑϨʔϜʹม͢Δ
্ख͍͔ͣ͘ ࣌ؒΛ࿘අ
WEBͷใ͔Β֤ฑͷಛΛநग़ ςΩετղੳ ΟεΩʔͰػցֶशʹઓ ̍ʣWebαΠτ͔Β֤ฑͷઆ໌ίϝϯτΛεΫϨΠϐϯά
WEBͷใ͔Β֤ฑͷಛΛநग़ ςΩετղੳ ΟεΩʔͰػցֶशʹઓ ̍ʣWebαΠτ͔Β֤ฑͷઆ໌ίϝϯτΛεΫϨΠϐϯά
WEBͷใ͔Β֤ฑͷಛΛநग़ ςΩετղੳ ΟεΩʔͰػցֶशʹઓ ̍ʣWebαΠτ͔Β֤ฑͷઆ໌ίϝϯτΛεΫϨΠϐϯά ฑͷղઆจ ޱίϛจ ͳΜͱ͔εΫϨΠϐϯάྃ ख࡞ۀͰCSV࡞
WEBͷใ͔Β֤ฑͷಛΛநग़ ςΩετղੳ ΟεΩʔͰػցֶशʹઓ ̎ʣmecabͰ୯ޠʹׂ͠ɺܗ༰ࢺͱ෭ࢺ͚ͩऔΓग़͢ ߳Γڧ͍Ͱ͕͢ɺຯ͖ͬ͢Γ͍ͯͯ͠ɺඒຯ͍͠Ͱ͢ɻ ߳Γ//ڧ͍/Ͱ͢/͕/ɺ/ຯ//͖ͬ͢Γ/ͯ͠/͍ͯ/ɺ /ඒຯ͍͠/Ͱ͢/ɻ
WEBͷใ͔Β֤ฑͷಛΛநग़ ςΩετղੳ ΟεΩʔͰػցֶशʹઓ ̏ʣग़ݱͨ͠୯ޠΛ·ͱΊɺTFIDFܭࢉͰಛతͳޠΛಛఆ BoW(Bag-of-Words) ܗࣜʹม จॻ ߦྻԽ TFIDF ܭࢉ
WEBͷใ͔Β֤ฑͷಛΛநग़ ςΩετղੳ ΟεΩʔͰػցֶशʹઓ ̏ʣग़ݱͨ͠ܗ༰ࢺΛ·ͱΊɺTFIDFܭࢉͰಛతͳޠΛಛఆ TFIDF=TF×IDF TF→Ұ࿈ͷจॻͷͦͷ୯ޠͷग़ݱස IDF→ʢͦͷ୯ޠ͕͋Δจॻͷʗશจॻʣͷٯ ʲͦͷจॻΒ͠͞ʳ ɹͦͷจॻͰଟ͘Ͱͯ͘ΔͷʹɺଞͷจॻͰ͋·ΓͰͯ͜ͳ͍୯ޠ BoW(Bag-of-Words)ܗࣜʹม
จॻߦྻԽ TFIDFܭࢉ BoWܗࣜ=จॻΛͱͯ͠ѻ͏ʢϕΫτϧԽ͢Δʣ ※ࠓճTFʢ୯ޠͷग़ݱසʣΛ༻
WEBͷใ͔Β֤ฑͷಛΛநग़ ςΩετղੳ ΟεΩʔͰػցֶशʹઓ ̐ʣ୯ޠΛͬͯࣅ͍ͯΔฑΛΫϥελϦϯά ֤ฑͷಛతͳ୯ޠTOP5Λͬͯ5ͭʹΫϥελϦϯά …͠Α͏ͱࢥͬͨΒ ˎ൱ఆʮͳ͍ʯͷߟྀ ˎআ֎ͨ͠΄͕͍͍୯ޠ ɹɹɹɹɹɹɹɹɹ ߟྀෆͰɺ
ͳ͍ΫϥελϦϯάʹ
͖ͳฑ͔ΒͨͿΜ͓ޱʹ߹͏ͷΛਪન Ϩίϝϯυ ΟεΩʔͰػցֶशʹઓ ຊ·Ͱʹ౸ୡͰ͖ͣ…ɻ ڧௐϑΟϧλϦϯάͱ͔ͰΓ͔͚ͨͬͨͲɺ ·ͣσʔλΛूΊΔͱ͍͏࠷େͷน͕ͬͯΔɻ
ΟεΩʔͰػցֶशʹઓͷཱྀଓ͘… Ϙτϧͷϥϕϧ͔ΒฑΛผ ը૾ೝࣝ WEBͷใ͔Β֤ฑͷಛΛநग़ ςΩετղੳ ͖ͳฑ͔ΒͨͿΜ͓ޱʹ߹͏ͷΛਪન Ϩίϝϯυ ˎਫ਼্ʂ90%ਖ਼ղ͍ͨ͠ɻOpenCVͬͯΈ͍ͨ ˎσʔλ૿ྔͱ൱ఆޠআ֎બఆɺϊΠζͷѻ͍ͷߟྀ ˎ͜Ε͔ΒؤுΔʂϢʔβෳධՁͰڧௐϑΟϧλϦϯά
WebΞϓϦέʔγϣϯ
·ͱΊ ˎPytonistaʹಌΕͨҰհͷੳͨͿΜPyLv1.5͘Β͍ʹͳΕͨ ˎ࠷ॳͷӈࠨΘ͔Βͳ͍࣌ɺؒͱҰॹ͕͍͍ ˎࣗͷڵຯͱֻ͚߹ΘͤΔͷͬͯؤுΓ͕࣋ଓ͢Δ ˎσʔλूΊΔͷେมͬͯᷚຊͩͬͨ ˎ·͡ͰwebΞϓϦέʔγϣϯΘ͔Γ·ͤΜ ɹಘҙͳਓɺͪΐͬͱͬͯΔਓɺ͓༑ୡʹͳ͍ͬͯͩ͘͞
͋Γ͕ͱ͏͍͟͝·ͨ͠ɻ